GroundWork adds validation, magic wand and other features to speed up image classification for machine learning
Information from GroundWork
In April, Azavea made its imagery segmentation and classification tools, GroundWork, freely available to anyone to label geospatial imagery for machine learning models. The tool has been updated with new features including validation, the introduction of a “magic wand” to speed up labelling, and a global label view and clickable task map.
Validation gives users the ability to review their or their team members’ work and approve or edit their labels. This is an optional way to double-check the accuracy of a training dataset, and the owner of a project can grant validation privileges (or revoke them) for any person they invite to collaborate on a project. Validators can access previously labelled tasks from a project’s home page and see who worked on the task originally, as well as when it was completed.
The semantic segmentation labelling workflow has been enhanced with a new “Magic Wand” tool, which allows users to mass select similar pixels in an image. Two new features make it easier to inspect and amendment projects directly from the project overview page: the global label view lets users filter according to labels right on the project overview page, while the clickable task map allows them to click on a grid cell and opens it directly in the task editor.